UD for Sanskrit 
Tokenization and Word Segmentation
- In general, words are delimited by whitespace characters. Description of exceptions follows.
- If a punctuation mark is attached to a neighboring word, we tokenize it as a separate token (word).
- Compounds are split and sandhi undone as it is the custom in Sanskrit linguistics. Sometimes it is necessary because one or more parts of an orthographic word is syntactically independent of the others. However, even parts of a nominal compound may be separated in order to make compound splitting consistent and predictable. Occasionally it is possible that parts of a multi-word token end up being connected with a compound relation.
Morphology
Tags
- Sanskrit uses 16 universal POS categories; the current data does not contain any occurrences of the SYM category.
- The only auxiliary verb is the copula अस् (as) “to be”.
Nominal Features
- Nominal words (NOUN, PROPN and PRON) have an inherent Gender feature with values
Masc,FemorNeut. - Number has 3 possible values:
Sing,DualandPlur. - Case has 8 possible values:
Nom,Gen,Dat,Acc,Loc,Abl,Ins,Voc. It occurs with the nominal words, i.e., NOUN, PROPN, PRON, ADJ, DET, NUM. It also occurs with some non-finite verb forms such as participles.
Degree and Polarity
Verbal Features
- Aspect is marked if it is perfective (
Perf). - Mood has 4 possible values:
Ind,Imp,Jus,Opt. - Voice has 4 possible values:
Act,Pass,Mid,Caus.
Pronouns, Determiners, Quantifiers
- PronType is used with pronouns (PRON) and determiners (DET).
- Person is a lexical feature of personal pronouns (PRON) and has three values,
1,2and3.
Other Features
- There is one language-specific feature:
Syntax
Core Arguments, Oblique Arguments and Adjuncts
- Nominal subject (nsubj) is a noun phrase in the nominative case, without preposition.
- For the purpose of UD the objects are divided to core objects, labeled obj or iobj,
and oblique objects, labeled obl:arg.
- Bare accusative objects are considered core.
- All prepositional objects are considered oblique.
- In passive clauses, the subject is labeled with nsubj:pass or csubj:pass, respectively.
Non-verbal Clauses
- Clauses with nominal predicates either use the copula अस् (as) “to be”, or they use simple juxtaposition without copula.
Relations Overview
- The following relation subtypes are used in Sanskrit:
- acl:relcl for relative adnominal clauses
- nsubj:cop for subjects of nominal clauses (with or without copula)
- nsubj:pass for nominal subjects of passive verbs
- csubj:pass for clausal subjects of passive verbs
- obl:agent for agents of passive verbs
- obl:arg for prepositional objects
- nmod:poss for possessive/genitive modifiers
- nummod:gov for cardinal numbers that are attached as children of the counted noun but govern its case
- advmod:emph for adverbs or particles that modify noun phrases and emphasize or negate them
Treebanks
There are two Sanskrit UD treebanks: